Photorealistic asks for something that looks like a photograph, flaws included; hyperrealistic asks for a heightened illusion, more than the human eye could see, so it tends to come out glossy. Realistic, cinematic and anamorphic get mixed up too, and each is covered below.
In Wikipedia’s words, hyperrealist work uses “additional, often subtle, pictorial elements to create the illusion of a reality which in fact either does not exist or cannot be seen by the human eye.”1
- Hyperrealistic means more than real, not more accurate. It is the single most misused word in image prompting.
- Photorealistic means "like a photograph" — including a photograph's flaws, which is what most people actually want.
- Realism is the loosest of the three and mostly means "not stylised". It is the weakest instruction of the five.
- Cinematic is not a look, it is a bundle — aspect ratio, lighting, colour and lens all at once. Say which part you mean.
- Anamorphic is a physical lens with specific, checkable side effects: oval bokeh, horizontal flares, wide frame.
- Upscaling is not one of these. It changes resolution; it does not change how real something looks.
- “8K”, “masterpiece” and “highly detailed” are not settings. The image size comes from the tool’s settings. See quality words.
"Photorealistic" asks for a photograph. "Hyperrealistic" asks for something a camera could never take.
If you type the second one hoping for the first, you get exactly what you asked for — and it looks fake.
The three realism words
Photorealism — like a photograph
Photorealism began as an art movement among American painters in the late 1960s and early 1970s: work that studies a photograph and reproduces it as faithfully as possible in another medium. Its defining discipline is restraint — strict photorealist painters often left out emotion, politics and storytelling, aiming at what the photograph recorded rather than what they felt about it.2
That is the useful sense for a prompt. A photograph has limits — one exposure, one focal plane, grain, an unflattering moment. Asking for photorealism is asking for those limits, and the limits are what make it read as real.
Hyperrealism — beyond what a photograph shows
Hyperrealism emerged separately from the early 1970s; the name comes from a 1973 exhibition in Brussels. It resembles a high-resolution photograph and is seen as an advancement of photorealism — it aims at the illusion of a reality that does not exist, or that the human eye cannot see. Unlike photorealism, it is often narrative and emotional.3
So the word does not mean "extra realistic". It means heightened. Every surface sharper than a lens resolves, every texture more present than it would be, nothing lost to depth of field. In a prompt it reliably produces more detail, more contrast, more everything — the exact over-processed look people are usually trying to escape.
Use it deliberately or not at all. It is the right word for a poster, a product hero shot, a piece of concept art. It is the wrong word for anything meant to pass as a photograph.
Realism — the weak one
Realism is the broadest of the three and carries the least instruction. In practice it tells a model roughly "not stylised, not cartoon, not illustration" — a negative rather than a direction.
It is not useless, it is just imprecise. If you are steering away from an illustrative default it will help. If you want a specific kind of image, one of the other two says more.
Hyperrealistic is not a stronger photorealistic. They are different destinations. Typing the first while wanting the second is the most common self-inflicted wound in image prompting — see why AI output looks fake for why the model is already leaning that way without your help.
Cinematic is a bundle, not a look
Say which part you actually want
"Cinematic" is the vaguest useful word in prompting. It carries a whole package, and the model picks which parts to apply — usually the most obvious ones.
Unpacked, it is roughly four separate instructions:
Naming the part beats naming the bundle. "2.39:1, single hard key from frame left, teal-and-amber grade, shallow focus" is a different quality of instruction from "cinematic", and it is repeatable — you can change one element without losing the others.
The failure mode of "cinematic" is getting the grade and nothing else — a flatly-lit, centred, ordinary image with a blue tint on it.
Anamorphic is a real object with real side effects
What the lens physically does
Unlike the other four words, this one names a piece of equipment, so its effects are specific and checkable.
An anamorphic lens squeezes the image horizontally onto the sensor and it is stretched back out afterwards, capturing a wider field than a spherical lens on the same film. The classic squeeze is 2x, designed to give roughly a 2.39:1 frame on 35mm.4
Three consequences you can ask for by name:
Naming the effects works better than naming the lens, because the effects are visible in stills and the word is an inference from them. And a warning worth having: anamorphic flares are the single most overused signifier of "expensive" — a blue streak across a frame that has no bright light in it reads as costume, not cinematography.
Quality words: ultra realistic, 8K, highly detailed, masterpiece
The plain point: typing “8K” does not make an 8K image, and words like “masterpiece” or “highly detailed” are not settings. The size of a picture comes from a setting in the tool, or from the model’s fixed sizes. To make a picture look more real, describe real things: the light, the lens and the surface.
What the image makers’ own guides say
Of the big vendors’ guides checked here, only Google’s Imagen guide recommends quality words. That guide sits in Vertex AI documentation which Google now marks as no longer being updated. It says “Certain keywords can let the model know that you’re looking for a high-quality asset” and lists “4K, HDR, Studio Photo” for photos, with examples made on Imagen 3, an older model.6
The others point somewhere else. Black Forest Labs, maker of FLUX.2, asks for camera details: “For photorealism, specify camera models, lenses, and film stocks.”7 OpenAI says to “Request ‘photorealistic’ or ‘real photograph’ explicitly when that is the goal”, and to set size and quality as settings, not words.8 Midjourney says “Short and simple prompts typically generate the best images” and warns against long lists.9 Its Raw mode switches off Midjourney’s own styling; with simple prompts, it says, “you’ll often get more realistic, photo-like images.”10 Hugging Face’s prompting guide for its diffusers library, which runs Stable Diffusion and other open models, says to write “a structured narrative, not a keyword list.”11
Apart from Google’s Imagen guide, none of the vendor guides cited in this section recommends “8K”, “masterpiece” or “highly detailed” as prompt words.Reasoning, 13 Oct 2026 — the OpenAI, Black Forest Labs, Gemini API, Midjourney Prompt Basics and Hugging Face prompting pages cited in this section were read on 13 Oct 2026. None of them recommends “8K”, “masterpiece” or “highly detailed” as prompt words; the Gemini page uses “4K” only as an output size.
Tags like “masterpiece” come from community Stable Diffusion models. Some of them were trained on captions written as lists of tags, and their makers ask for quality tags. The card for one such model, Animagine XL 4.0, says it “was trained with tag-based captions” and suggests ending a prompt with “masterpiece, high score, great score, absurdres”.12 That is community practice for that kind of model. Use it where the model card asks for it, and do not carry it over to tools whose own guides do not.
| Word | What people think it does | What it actually does | Write this instead |
|---|---|---|---|
| 8K, 4K | People think: it makes a bigger, sharper file. | Actually: it does not set the size (see the next table). Google’s Imagen guide lists “4K” as a quality keyword, so on some models it may nudge the look. | Instead: pick the size in the tool’s settings, and upscale afterwards if you need more. |
| Ultra realistic, hyperrealistic | People think: more real. | Actually: in art, hyperrealism means heightened, not truer (see above). OpenAI’s guide asks for “photorealistic” or “real photograph” instead. | Instead: “real photograph”, plus a camera, a lens and a light source. In Midjourney, try --raw. |
| Highly detailed, ultra-detailed | People think: more detail everywhere. | Actually: the vendor guides above do not ask for it (Google’s Imagen guide lists plain “detailed” for art and illustration). In this site’s use it tends toward an evenly sharpened, over-processed look (see why AI output looks fake). | Instead: name the detail: visible pores, a frayed cuff, dust on a shelf. More lines on micro detail prompts. |
| Masterpiece, best quality | People think: a better picture. | Actually: a training tag on some community Stable Diffusion models. The vendor guides above do not use it. | Instead: on those models, follow the model card. Elsewhere, describe the scene. |
| Photorealistic | People think: the same as ultra realistic. | Actually: the word OpenAI’s guide suggests when you want a photo. | Instead: keep it, and add the flaws a real photo has: grain, one light, a slightly unflattering moment. |
Does “8K” in a prompt set the size?
No. Every tool below sets size with a setting or a fixed size, and none of them documents 8K.Reasoning, 13 Oct 2026 — read from the five size rows above: each sets size with a setting or a fixed size; the largest size any of them names is 4K (Gemini and OpenAI), and the diffusers page names no maximum.
| Tool | Where the size is set | Largest size its docs list |
|---|---|---|
| Gemini (Nano Banana) | Setting: image_size, with an uppercase K. 1K is the default. | Largest: 4K.13 |
| OpenAI GPT Image (API) | Setting: size, for example 1536x1024. | Largest: no edge over 3840 pixels, and no more pixels in all than OpenAI’s 4K figure.14 |
| FLUX.2 | Setting: the output dimensions you ask for. | Largest: 4 megapixels, for example 2048×2048.7 |
| Midjourney V8.2 | Setting: standard (SD) or HD, then the upscalers. | Largest: 2048 x 2048 pixels for a square image; other shapes change the numbers.15 |
| Stable Diffusion (diffusers) | Setting: height and width, in pixels. | Largest: depends on the model and your hardware.16 |
If you need a bigger file than the tool makes, upscale afterwards, and know what that does: AI upscaling explains it.
Size is a setting, not a word. For realism, describe the camera, the light and the surface instead of praising the picture. Keep quality tags for the community models whose cards ask for them.
Upscaling is a different axis entirely
People reach for upscaling when an image looks unconvincing, and it will not help. Upscaling changes resolution and can add plausible detail. It does not change lighting, composition, or whether the thing looks real.
What it is good for: rescuing a composition you like from a low-resolution generation, preparing something for print, and adding detail to a small crop (the new detail is a guess, not a recovery). What it will not do is fix "this looks fake" — that is a lighting and surface problem, and enlarging it makes the wrongness bigger.
Fix the image, then upscale. Prompt libraries for the second step are on ultra-detailed upscale prompts; the diagnosis for the first is on why your generation came out wrong. What an upscaler invents, and when that matters, is on upscaling myths.
Photorealistic — like a photograph, flaws included.
Hyperrealistic — sharper and more present than a photograph, deliberately unreal.
Realistic — not stylised. Weak instruction.
Cinematic — a bundle of frame, light, colour and lens. Name the part.
Anamorphic — a specific lens: oval bokeh, horizontal flares, wide frame.
Upscaled — bigger, not better. A resolution step, not a realism one.
8K, masterpiece — not settings. Pick the size in the tool; describe the scene.
Choosing the word
Questions people ask
Does 8K in a prompt make the image 8K?
No. The size comes from the tool’s settings: Gemini image models, for example, “generate 1K images by default” and need the image_size setting for 2K or 4K.13 None of the tools in the size table above documents 8K, so a bigger file usually means upscaling afterwards.
Hyperrealistic or photorealistic: which should I use?
Use photorealistic when the picture should pass as a photo; OpenAI’s guide says to “Request ‘photorealistic’ or ‘real photograph’ explicitly when that is the goal.”8 Use hyperrealistic only for a heightened, poster-like look: in art, it aims at “the illusion of a reality which in fact either does not exist or cannot be seen by the human eye.”1
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What does typing “8K” in an image prompt do to the size of the picture?
- It makes the picture 8K wide
- It does not set the size; a setting in the tool does
- It doubles the size of the picture
Show the answer
Answer: It does not set the size; a setting in the tool does. The page says the size comes from a setting in the tool or from the model’s fixed sizes, and none of the tools it lists documents 8K.
2. How is hyperrealistic different from photorealistic?
- Hyperrealistic is simply a stronger photorealistic
- They mean the same thing to an image model
- Photorealistic asks for a photo; hyperrealistic asks for a heightened look
Show the answer
Answer: Photorealistic asks for a photo; hyperrealistic asks for a heightened look. The page says photorealistic asks for a photograph, flaws included, while hyperrealistic means heightened, more than a camera would show.
3. What does the page suggest writing instead of “highly detailed”?
- Name the detail you want, like visible pores or a frayed cuff
- Add “masterpiece” and “best quality” after it
- Repeat “highly detailed” a few more times
Show the answer
Answer: Name the detail you want, like visible pores or a frayed cuff. The page says to name the actual detail you want, because the vendor guides it quotes do not ask for “highly detailed”.
Related on this site: Art Styles for AI Prompts · Why Do AI Images Look Plastic?
- 1Wikipedia, Hyperrealism (visual arts), read at source 13 Oct 2026hyperrealist work uses “additional, often subtle, pictorial elements to create the illusion of a reality which in fact either does not exist or cannot be seen by the human eye.” ↑ahyperrealist work uses “additional, often subtle, pictorial elements to create the illusion of a reality which in fact either does not exist or cannot be seen by the human eye.” ↑b
- 2Read at source 11 Sep 2026. Photorealism — "a genre of art… in which an artist studies a photograph and then attempts to reproduce the image as realistically as possible in another medium", beginning "in the late 1960s and early 1970s in the United States"; the term was "coined by Louis K. Meisel in 1969". The restraint claim is not on that page: it comes from Hyperrealism, which says “Strict Photorealist painters tended to imitate photographic images” and “They often omitted human emotion, political value, and narrative elements.” Checked 6 Sep 2026, re-read 11 Sep 2026 and 13 Oct 2026. Corrected 13 Oct 2026this line used to quote “tended to omit human emotion, political value, and narrative elements”, which is not the page’s wording, and the article said the painters “deliberately” kept emotion out. ↑
- 3Read at source 11 Sep 2026: Hyperrealism (visual arts) — "a genre of painting and sculpture resembling a high-resolution photograph"; "The term is primarily applied to an independent art movement and art style in the United States and Europe that has developed since the early 1970s"; "The art dealer Isy Brachot coined the French word hyperréalisme, meaning hyperrealism, as the title of a major exhibition and catalogue at his gallery in Brussels in 1973"; "Hyperrealism is considered an advancement of photorealism by the methods used to create the resulting paintings or sculptures"; hyperrealist work "often, unlike photorealism, is narrative and emotive in its depictions" and uses "additional, often subtle, pictorial elements to create the illusion of a reality which in fact either does not exist or cannot be seen by the human eye". Checked 6 Sep 2026, re-read 11 Sep 2026 and 13 Oct 2026. Corrected 13 Oct 2026the article said hyperrealism is “described as so hyperreal that it becomes unreal” and “deliberately” adds narrative; the page says neither. ↑
- 4Read at source 11 Sep 2026. Anamorphic format — "Anamorphic capture employs cylindrical lenses to horizontally compress or ‘squeeze’ the image during recording." and "A complementary lens is then used during projection to expand the image back"; "the anamorphic lenses in virtually all 35 mm anamorphic systems provide a 2:1 squeeze"; it has "an inherently wider horizontal field of view", and "This has led to the common claim that anamorphic lenses have shallower focus"; out-of-focus points of light appear "as vertical ovals rather than circles"; the signature flare is "a long horizontal line, usually with a blue tint". The bokeh proportion is from Cooke Optics: "with 2x anamorphics the resulting oval shape will be 50% as wide as it is tall." The 2.39:1 pairing is from Desqueezed: "On Super 35 sensors, 2x squeeze was historically paired with Academy ratio (1.375:1) gates to produce 2.39:1 Cinemascope." Checked 6 Sep 2026, re-read 11 Sep 2026 and 13 Oct 2026.↑
- 5Cooke Optics, Distorting reality - What are anamorphic lenses?, read at source 23 Sep 2026, re-read 13 Oct 2026“with 2x anamorphics the resulting oval shape will be 50% as wide as it is tall.” ↑
- 6Google Cloud, Prompt and image attribute guide, read at source 13 Oct 2026 under the banner “Vertex AI documentation is no longer being updated”: “Certain keywords can let the model know that you're looking for a high-quality asset.” and “Photos - 4K, HDR, Studio Photo”. The page also lists “detailed” among its modifiers for art and illustration, and notes that its quality-modifier example images were made with Imagen 3.↑
- 7Black Forest Labs, Prompting Guide — FLUX.2 [pro] & [max], read at source 13 Oct 2026“For photorealism, specify camera models, lenses, and film stocks.” and “‘Shot on Fujifilm X-T5, 35mm f/1.4’ produces more authentic results than just ‘professional photo.’” ↑a“Resolution limits: Minimum 64×64, maximum 4MP (e.g., 2048×2048).” ↑b
- 8OpenAI, Image prompting, read at source 13 Oct 2026“Request ‘photorealistic’ or ‘real photograph’ explicitly when that is the goal.” and “Set API parameters separately from the prompt.” ↑a“Request ‘photorealistic’ or ‘real photograph’ explicitly when that is the goal.” and “Set API parameters separately from the prompt.” ↑b
- 9Midjourney, Prompt Basics, re-read directly at source 13 Oct 2026“Short and simple prompts typically generate the best images with Midjourney.” and “Avoid making long lists or detailed instructions; these can confuse the process.” ↑
- 10Midjourney, Raw, re-read directly at source 13 Oct 2026“With simple prompts, you’ll often get more realistic, photo-like images.” ↑
- 11Hugging Face, Diffusers: Prompting, read at source 13 Oct 2026“Use these elements as a structured narrative, not a keyword list. Modern models understand language better than keyword matching.” ↑
- 12Cagliostro Research Lab, Animagine XL 4.0 model card (Hugging Face), read at source 13 Oct 2026“The model was trained with tag-based captions and the tag-ordering method.” and, for the end of a prompt, “masterpiece, high score, great score, absurdres”. ↑
- 13Google, Nano Banana image generation (Gemini API), read at source 13 Oct 2026“Gemini 3 image models generate 1K images by default”; “To generate higher resolution assets, specify the
image_sizein theresponse_format.” and “You must use an uppercase ‘K’”, with 1K, 2K and 4K as the examples. ↑a“Gemini 3 image models generate 1K images by default”; “To generate higher resolution assets, specify theimage_sizein theresponse_format.” and “You must use an uppercase ‘K’”, with 1K, 2K and 4K as the examples. ↑b - 14OpenAI, Image generation (API guide), read at source 13 Oct 2026“Width and height must be multiples of 16, the aspect ratio must be between 1:3 and 3:1, and neither edge may exceed 3840 pixels.” and “The total pixel count must be between 655,360 and 8,294,400 (4K).” ↑
- 15Midjourney, Upscalers, re-read directly at source 13 Oct 2026“With Midjourney V8.2, your SD images start at 1024 x 1024px (using the default aspect ratio). You can then use the Creative and Subtle upscalers to bump up your images to 2048 x 2048px.” and “HD images generate at 2048 x 2048px automatically, and currently cannot be upscaled further.” ↑
- 16Hugging Face, Diffusers: Text-to-image (StableDiffusionPipeline), read at source 13 Oct 2026
height“The height in pixels of the generated image.” andwidth“The width in pixels of the generated image.” ↑
Photorealism and Hyperrealism as art movements — origins, dates and the distinction between them, including hyperrealism's 1973 Brussels naming · anamorphic optics — squeeze factor, oval bokeh geometry and horizontal flare characteristics. Checked 6 September 2026; every source above re-opened and quoted 11 September 2026. Re-checked in full 13 October 2026: every source re-opened; two quotes had drifted from the pages’ wording and two sentences said more than their source, all corrected (see sources 1 to 3). The quality-words section was added the same day from the vendors’ own guides.
The art history and the optics are sourced, and so is what each vendor’s guide says about quality words. What each word does inside a given model is not. Models weight these terms differently, and the same word can behave differently between versions of the same model. The definitions are stable; the model responses are not — test the pair that matters to you rather than trusting the effect described here.
The through-line: these words are not a scale from slightly real to very real. They point at different places, and hyperrealistic points away from the photograph most people are asking for.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK